modelfusion 0.115.0 → 0.116.1

package/CHANGELOG.md

@@ -1,5 +1,51 @@
 # Changelog
 
+## v0.116.0 - 2024-01-05
+
+### Added
+
+- Semantic classifier. An easy way to determine a class of a text using embeddings. Example:
+
+  ```ts
+  import { SemanticClassifier, openai } from "modelfusion";
+
+  const classifier = new SemanticClassifier({
+    embeddingModel: openai.TextEmbedder({
+      model: "text-embedding-ada-002",
+    }),
+    similarityThreshold: 0.82,
+    clusters: [
+      {
+        name: "politics" as const,
+        values: [
+          "isn't politics the best thing ever",
+          "why don't you tell me about your political opinions",
+          "don't you just love the president",
+          "don't you just hate the president",
+          "they're going to destroy this country!",
+          "they will save the country!",
+        ],
+      },
+      {
+        name: "chitchat" as const,
+        values: [
+          "how's the weather today?",
+          "how are things going?",
+          "lovely weather today",
+          "the weather is horrendous",
+          "let's go to the chippy",
+        ],
+      },
+    ],
+  });
+
+  console.log(await classifier.classify("don't you love politics?")); // politics
+  console.log(await classifier.classify("how's the weather today?")); // chitchat
+  console.log(
+    await classifier.classify("I'm interested in learning about llama 2")
+  ); // null
+  ```
+
 ## v0.115.0 - 2024-01-05
 
 ### Removed

package/classifier/SemanticClassifier.cjs

@@ -0,0 +1,75 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.SemanticClassifier = void 0;
+const embed_js_1 = require("../model-function/embed/embed.cjs");
+const cosineSimilarity_js_1 = require("../util/cosineSimilarity.cjs");
+class SemanticClassifier {
+    constructor({ clusters, embeddingModel, similarityThreshold, }) {
+        Object.defineProperty(this, "clusters", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "embeddingModel", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "similarityThreshold", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "embeddings", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        this.clusters = clusters;
+        this.embeddingModel = embeddingModel;
+        this.similarityThreshold = similarityThreshold;
+    }
+    async getEmbeddings() {
+        if (this.embeddings != null) {
+            return this.embeddings;
+        }
+        const embeddings = [];
+        for (const cluster of this.clusters) {
+            const clusterEmbeddings = await (0, embed_js_1.embedMany)(this.embeddingModel, cluster.values);
+            for (let i = 0; i < clusterEmbeddings.length; i++) {
+                embeddings.push({
+                    embedding: clusterEmbeddings[i],
+                    clusterValue: cluster.values[i],
+                    clusterName: cluster.name,
+                });
+            }
+        }
+        this.embeddings = embeddings; // lazy caching
+        return embeddings;
+    }
+    async classify(value) {
+        const valueEmbedding = await (0, embed_js_1.embed)(this.embeddingModel, value);
+        const clusterEmbeddings = await this.getEmbeddings();
+        const allMatches = [];
+        for (const embedding of clusterEmbeddings) {
+            const similarity = (0, cosineSimilarity_js_1.cosineSimilarity)(valueEmbedding, embedding.embedding);
+            if (similarity >= this.similarityThreshold) {
+                allMatches.push({
+                    similarity,
+                    clusterValue: embedding.clusterValue,
+                    clusterName: embedding.clusterName,
+                });
+            }
+        }
+        // sort (highest similarity first)
+        allMatches.sort((a, b) => b.similarity - a.similarity);
+        return allMatches.length > 0
+            ? allMatches[0].clusterName
+            : null;
+    }
+}
+exports.SemanticClassifier = SemanticClassifier;

package/classifier/SemanticClassifier.d.ts

@@ -0,0 +1,25 @@
+import { Vector } from "../core/Vector.js";
+import { EmbeddingModel } from "../model-function/embed/EmbeddingModel.js";
+export interface SemanticCluster<VALUE, NAME extends string> {
+    name: NAME;
+    values: VALUE[];
+}
+export declare class SemanticClassifier<VALUE, CLUSTERS extends Array<SemanticCluster<VALUE, string>>> {
+    readonly clusters: CLUSTERS;
+    readonly embeddingModel: EmbeddingModel<VALUE>;
+    readonly similarityThreshold: number;
+    private embeddings;
+    constructor({ clusters, embeddingModel, similarityThreshold, }: {
+        clusters: CLUSTERS;
+        embeddingModel: EmbeddingModel<VALUE>;
+        similarityThreshold: number;
+    });
+    getEmbeddings(): Promise<{
+        embedding: Vector;
+        clusterValue: VALUE;
+        clusterName: string;
+    }[]>;
+    classify(value: VALUE): Promise<ClusterNames<CLUSTERS> | null>;
+}
+type ClusterNames<CLUSTERS> = CLUSTERS extends Array<SemanticCluster<unknown, infer NAME>> ? NAME : never;
+export {};

package/classifier/SemanticClassifier.js

@@ -0,0 +1,71 @@
+import { embed, embedMany } from "../model-function/embed/embed.js";
+import { cosineSimilarity } from "../util/cosineSimilarity.js";
+export class SemanticClassifier {
+    constructor({ clusters, embeddingModel, similarityThreshold, }) {
+        Object.defineProperty(this, "clusters", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "embeddingModel", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "similarityThreshold", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        Object.defineProperty(this, "embeddings", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        this.clusters = clusters;
+        this.embeddingModel = embeddingModel;
+        this.similarityThreshold = similarityThreshold;
+    }
+    async getEmbeddings() {
+        if (this.embeddings != null) {
+            return this.embeddings;
+        }
+        const embeddings = [];
+        for (const cluster of this.clusters) {
+            const clusterEmbeddings = await embedMany(this.embeddingModel, cluster.values);
+            for (let i = 0; i < clusterEmbeddings.length; i++) {
+                embeddings.push({
+                    embedding: clusterEmbeddings[i],
+                    clusterValue: cluster.values[i],
+                    clusterName: cluster.name,
+                });
+            }
+        }
+        this.embeddings = embeddings; // lazy caching
+        return embeddings;
+    }
+    async classify(value) {
+        const valueEmbedding = await embed(this.embeddingModel, value);
+        const clusterEmbeddings = await this.getEmbeddings();
+        const allMatches = [];
+        for (const embedding of clusterEmbeddings) {
+            const similarity = cosineSimilarity(valueEmbedding, embedding.embedding);
+            if (similarity >= this.similarityThreshold) {
+                allMatches.push({
+                    similarity,
+                    clusterValue: embedding.clusterValue,
+                    clusterName: embedding.clusterName,
+                });
+            }
+        }
+        // sort (highest similarity first)
+        allMatches.sort((a, b) => b.similarity - a.similarity);
+        return allMatches.length > 0
+            ? allMatches[0].clusterName
+            : null;
+    }
+}

package/classifier/index.cjs

@@ -0,0 +1,17 @@
+"use strict";
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __exportStar = (this && this.__exportStar) || function(m, exports) {
+    for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+__exportStar(require("./SemanticClassifier.cjs"), exports);

package/classifier/index.d.ts

@@ -0,0 +1 @@
+export * from "./SemanticClassifier.js";

package/classifier/index.js

@@ -0,0 +1 @@
+export * from "./SemanticClassifier.js";

package/index.cjs

@@ -14,6 +14,7 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
     for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
 };
 Object.defineProperty(exports, "__esModule", { value: true });
+__exportStar(require("./classifier/index.cjs"), exports);
 __exportStar(require("./core/index.cjs"), exports);
 __exportStar(require("./model-function/index.cjs"), exports);
 __exportStar(require("./model-provider/index.cjs"), exports);

package/index.d.ts

@@ -1,3 +1,4 @@
+export * from "./classifier/index.js";
 export * from "./core/index.js";
 export * from "./model-function/index.js";
 export * from "./model-provider/index.js";

package/index.js

@@ -1,3 +1,4 @@
+export * from "./classifier/index.js";
 export * from "./core/index.js";
 export * from "./model-function/index.js";
 export * from "./model-provider/index.js";

package/model-provider/mistral/MistralTextEmbeddingModel.d.ts

@@ -32,16 +32,16 @@ export declare class MistralTextEmbeddingModel extends AbstractModel<MistralText
     doEmbedValues(texts: string[], options: FunctionCallOptions): Promise<{
         response: {
             object: string;
-            data: {
-                object: string;
-                embedding: number[];
-                index: number;
-            }[];
             model: string;
             usage: {
                 prompt_tokens: number;
                 total_tokens: number;
             };
+            data: {
+                object: string;
+                embedding: number[];
+                index: number;
+            }[];
             id: string;
         };
         embeddings: number[][];
@@ -77,29 +77,29 @@ declare const MistralTextEmbeddingResponseSchema: z.ZodObject<{
     }>;
 }, "strip", z.ZodTypeAny, {
     object: string;
-    data: {
-        object: string;
-        embedding: number[];
-        index: number;
-    }[];
     model: string;
     usage: {
         prompt_tokens: number;
         total_tokens: number;
     };
-    id: string;
-}, {
-    object: string;
     data: {
         object: string;
         embedding: number[];
         index: number;
     }[];
+    id: string;
+}, {
+    object: string;
     model: string;
     usage: {
         prompt_tokens: number;
         total_tokens: number;
     };
+    data: {
+        object: string;
+        embedding: number[];
+        index: number;
+    }[];
     id: string;
 }>;
 export type MistralTextEmbeddingResponse = z.infer<typeof MistralTextEmbeddingResponseSchema>;

package/model-provider/ollama/OllamaChatModel.d.ts

@@ -40,11 +40,11 @@ export declare class OllamaChatModel extends AbstractModel<OllamaChatModelSettin
     get settingsForEvent(): Partial<OllamaChatModelSettings>;
     doGenerateTexts(prompt: OllamaChatPrompt, options: FunctionCallOptions): Promise<{
         response: {
+            model: string;
             message: {
                 role: string;
                 content: string;
             };
-            model: string;
             done: true;
             created_at: string;
             total_duration: number;
@@ -61,11 +61,11 @@ export declare class OllamaChatModel extends AbstractModel<OllamaChatModelSettin
     }>;
     restoreGeneratedTexts(rawResponse: unknown): {
         response: {
+            model: string;
             message: {
                 role: string;
                 content: string;
             };
-            model: string;
             done: true;
             created_at: string;
             total_duration: number;
@@ -82,11 +82,11 @@ export declare class OllamaChatModel extends AbstractModel<OllamaChatModelSettin
     };
     private processTextGenerationResponse;
     doStreamText(prompt: OllamaChatPrompt, options: FunctionCallOptions): Promise<AsyncIterable<import("../../index.js").Delta<{
+        model: string;
         message: {
             role: string;
             content: string;
         };
-        model: string;
         done: false;
         created_at: string;
     } | {
@@ -141,11 +141,11 @@ declare const ollamaChatResponseSchema: z.ZodObject<{
     eval_count: z.ZodNumber;
     eval_duration: z.ZodNumber;
 }, "strip", z.ZodTypeAny, {
+    model: string;
     message: {
         role: string;
         content: string;
     };
-    model: string;
     done: true;
     created_at: string;
     total_duration: number;
@@ -155,11 +155,11 @@ declare const ollamaChatResponseSchema: z.ZodObject<{
     load_duration?: number | undefined;
     prompt_eval_duration?: number | undefined;
 }, {
+    model: string;
     message: {
         role: string;
         content: string;
     };
-    model: string;
     done: true;
     created_at: string;
     total_duration: number;
@@ -185,19 +185,19 @@ declare const ollamaChatStreamChunkSchema: z.ZodDiscriminatedUnion<"done", [z.Zo
         content: string;
     }>;
 }, "strip", z.ZodTypeAny, {
+    model: string;
     message: {
         role: string;
         content: string;
     };
-    model: string;
     done: false;
     created_at: string;
 }, {
+    model: string;
     message: {
         role: string;
         content: string;
     };
-    model: string;
     done: false;
     created_at: string;
 }>, z.ZodObject<{
@@ -247,11 +247,11 @@ export declare const OllamaChatResponseFormat: {
             requestBodyValues: unknown;
             response: Response;
         }) => Promise<{
+            model: string;
             message: {
                 role: string;
                 content: string;
             };
-            model: string;
             done: true;
             created_at: string;
             total_duration: number;
@@ -271,11 +271,11 @@ export declare const OllamaChatResponseFormat: {
         handler: ({ response }: {
             response: Response;
         }) => Promise<AsyncIterable<import("../../index.js").Delta<{
+            model: string;
             message: {
                 role: string;
                 content: string;
             };
-            model: string;
             done: false;
             created_at: string;
         } | {

package/model-provider/openai/OpenAITextEmbeddingModel.d.ts

@@ -52,16 +52,16 @@ export declare class OpenAITextEmbeddingModel extends AbstractModel<OpenAITextEm
     doEmbedValues(texts: string[], callOptions: FunctionCallOptions): Promise<{
         response: {
             object: "list";
-            data: {
-                object: "embedding";
-                embedding: number[];
-                index: number;
-            }[];
             model: string;
             usage: {
                 prompt_tokens: number;
                 total_tokens: number;
             };
+            data: {
+                object: "embedding";
+                embedding: number[];
+                index: number;
+            }[];
         };
         embeddings: number[][];
     }>;
@@ -95,28 +95,28 @@ declare const openAITextEmbeddingResponseSchema: z.ZodObject<{
     }>;
 }, "strip", z.ZodTypeAny, {
     object: "list";
-    data: {
-        object: "embedding";
-        embedding: number[];
-        index: number;
-    }[];
     model: string;
     usage: {
         prompt_tokens: number;
         total_tokens: number;
     };
-}, {
-    object: "list";
     data: {
         object: "embedding";
         embedding: number[];
         index: number;
     }[];
+}, {
+    object: "list";
     model: string;
     usage: {
         prompt_tokens: number;
         total_tokens: number;
     };
+    data: {
+        object: "embedding";
+        embedding: number[];
+        index: number;
+    }[];
 }>;
 export type OpenAITextEmbeddingResponse = z.infer<typeof openAITextEmbeddingResponseSchema>;
 export {};

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "The TypeScript library for building AI applications.",
-  "version": "0.115.0",
+  "version": "0.116.1",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [